F1: Capacity for Self-Directed Action Is State-Dependent

Integration

The neuroscience is established. This premise integrates results from multiple biological mechanism files (catecholamine-PFC dynamics, acute stress effects on executive function, controllability circuit, LC-NE adaptive gain) into a single engineering-relevant claim: the human's capacity to access their own evaluative processes -- what this work calls autonomy -- is not a stable trait but a state variable that fluctuates with neurobiological conditions.

In the terminology of this work, the neuroscience describes mechanisms of autonomy degradation -- the evaluative machinery going offline under catecholamine disruption. The agency/autonomy distinction is this work's framing, not the source papers' terminology.

Why this matters for the architecture

If capacity is state-dependent, then any system interacting with a human without estimating their state is interacting blind. During degraded states, the human cannot use advice, evaluate options, or exercise second-order governance. Interventions designed for a capable person fail -- and may backfire -- when delivered to a degraded person.

This premise is the foundation for the entire architecture: it is why state estimation is necessary (D4), why the objective function targets capacity rather than behavior (D2), and why the backfire regime exists (P1).

Evidence status

ESTABLISHED for the neuroscience. HYPOTHESIS for the formalization as a control variable. The specific application to "capacity for self-directed action" as a formal control variable appears novel. [UNCERTAIN on novelty]